Process of User-oriented interactive flooding-leading rain forecast system Chen Jing 1 Zhongwei Yan 2 Jiarui Han 3 Jiao Meiyan 4 1. Numerical Weather Prediction Center, CMA 2. RCE-TEA, Institute of Atmospheric Physics, Beijing 3. Research Center for Strategic Development, CMA THORPEX Asia, Kunming, 1 Nov 2012
Process of User-oriented interactive flooding-leading rain forecast system Chen Jing 1 Zhongwei Yan 2 Jiarui Han 3 Jiao Meiyan 4 1. Numerical Weather Prediction Center, CMA 2. RCE-TEA, Institute of Atmospheric Physics, Beijing 3. Research Center for Strategic Development, CMA. - PowerPoint PPT Presentation
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Process of User-oriented interactive
flooding-leading rain forecast system
Chen Jing1 Zhongwei Yan2 Jiarui Han3 Jiao Meiyan4
1. Numerical Weather Prediction Center, CMA
2. RCE-TEA, Institute of Atmospheric Physics, Beijing
3. Research Center for Strategic Development, CMA
THORPEX Asia, Kunming, 1 Nov 2012
limit of predictability
Why user-oriented?
The meteorological model, as a chaotic system, is of
limited predictability. General improvement of large-scale
forecast has, asymptotically, been limited.
However, for a given user, at a specified scale, there is
still great potential of improvement, especially in the
context of ensemble forecast.
User’s needs &decision-making
information
Conceptual User-oriented Interactive Forecast System
Meteorologicalforecast system
What to be user-oriented?
Key variable Initial condition with sensitive perturbations
114~121°E, 32~37°N, a 3°×3° grid-box Resolution 0.5°×0.5°
16
26
36
46
56
66
76
83% 86% 89% 92% 95% 98%
obgrandecmfbabj
0
5
10
15
20
25
30
53% 58% 63% 68% 73% 78% 83% 88%
obgrandecmfbabj
TIGGE Bias ---percentile distribution of the all TIGGE forecasts and
observations
0
50
100
150
200
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%百分位
(mm)
降雨量
ob
grand
ecmf
babj
If TIGGE forecast are accurate, the distribution of TIGGE forecasts and OB are almost the same. But there exists systematic forecast bias in all ensemble system, especially for more than 14.6mm. For this systematic bias, How to calibrate the bias ?
14.6mm
Distribution Calibration Method When samples size t was sufficiently large, precipitation observations on
user-end could form a distribution Ot, ,correspondingly precipitation
forecasts could also form a forecasts distribution Ft. . Because of
systematic forecast bias, on the same x percentile, forecast Pf was
different from observation Pob, that is Ft(x)≠ Ot(x).
If x<δ% , Pf > Pob , if x>δ% , Pf < Pob ; and if x=δ% , Pf = Pob
theoretically, precipitation observations (Ot) and precipitation
forecasts (Ft ) were identically distributed, Ft = Ot( Gneiting et al.,
2007) . That is, in the same x percentile, forecast Pf and
observation Pob should be the same.
Therefore, supposing (Ot) and precipitation forecasts (Ft ) were
identically distributed, let Ft(x)= Ot(x) in the same x percentile to
calibrate the forecast on user-end.
ETS verification results
Perfect score is 1; and 0 means no skill.
After calibration, forecasts improved.
BIAS Score
After calibration, all ensemble forecasts improved.
Perfect score is 1
Brier Score
0 is perfect score, and all ensemble
forecasts improved after calibration
User-oriented Interactive Forecasting System
Preliminary results
Dynamic Forecast Target——FLRT
FLRT in Regression method
The gap of water level
FLRT in Hydrological model method
FLRT in Regression method
FLRT in Hydrological model method
the dynamic FLRT reflect a change of flood-risk on user-end, but it ignored the low-risk cases, which is the different from the hydrological model. And it not only shows users to prevent high-flood-risk cases, but
provides a forecast target for forecast system (TIGGE).
The gap of water level to 27.5m
FLRT in Regression method
FLRT in Hydrological model method
1Jun.-31 Sep. 2008
FLRT v.s. TIGGE grand ensemble mean
FLRT in Regression method
FLRT in Hydrological model method
TIGGE ensemble mean
Although, there are several heavy rainfall events in 1Jun.-31 Sep. 2008, not every heavy rain could lead to a flood-risk. TIGGE ensemble mean could catch some
heavy rainfall events but not flood-leading events.
Flood Leading rain risk probabilistic forecast
TIGGE
Grand
Ensemble
( 162
member
s )
-30
0
30
60
0 20 40 60 80 100 1200
20
40
60
80
100
120
140
Days since June 1, 2008
Percen
tag
e o
f R
isk
Po
ssib
ilit
y
Members > Threshold All TIGGE Members
Precip
ita
tio
n T
hresh
old
(m
m)
the predicted probability of occurrence of the FLR events in Wangjiaba sub-basin, based on TIGGE grand ensemble forecast and
the dynamic FLRT with the user-end information.
Conclusion
Forecasting flood-leading rainfall at a specific user-scale is feasible with TIGGE data, as long as the ensemble products are